Modelpublic

brijeshvadi/eprocure-product-embeddings

sourceHugging Faceapache-2.0updated 6mo agoView on Hugging Face

0likes10downloads

Model Card

e-Procure Product Embeddings

Bilingual (English/Arabic) sentence embeddings fine-tuned for B2B procurement product matching on the e-Procure platform.

Model Description

Fine-tuned from sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2 on 48,000 product pairs from Saudi Arabian B2B procurement catalogs. Optimized for matching purchase requests to supplier catalog items across English and Arabic.

Key Capabilities

—Cross-lingual matching: Match English RFQ terms to Arabic product descriptions and vice versa
—Industry-specific: Trained on construction, electrical, HVAC, plumbing, and safety equipment catalogs
—SKU-aware: Understands product codes, part numbers, and technical specifications

Training Data

Category	English Pairs	Arabic Pairs	Cross-lingual
Construction Materials	8,200	6,100	3,400
Electrical Equipment	7,500	5,800	2,900
HVAC Systems	5,100	4,200	2,100
Plumbing Supplies	4,800	3,600	1,800
Safety Equipment	3,900	2,800	1,500

Usage

python

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("brijeshvadi/eprocure-product-embeddings")

queries = ["3-phase circuit breaker 400A", "قاطع دائرة ثلاثي الطور 400 أمبير"]
products = ["ABB SACE Tmax XT4 400A 3P MCCB", "Schneider NSX400N 3P 400A"]

query_emb = model.encode(queries)
product_emb = model.encode(products)

Architecture

—Base: paraphrase-multilingual-MiniLM-L12-v2
—Embedding Dim: 384
—Max Seq Length: 128
—Pooling: Mean pooling
—Training Loss: MultipleNegativesRankingLoss + CosineSimilarityLoss

Platform Context

Built for e-Procure, a B2B procurement platform serving Saudi Arabian construction and industrial supply chains. The platform uses Next.js 15, Strapi CMS, and Redux Toolkit Query.